Rule Base Simplification and Constrained Learning for Interpretability in TSK Neuro-Fuzzy Modelling

Int. J. Fuzzy Syst. Appl. Pub Date : 2020-04-01 DOI:10.4018/ijfsa.2020040102

Sharifa Rajab

{"title":"Rule Base Simplification and Constrained Learning for Interpretability in TSK Neuro-Fuzzy Modelling","authors":"Sharifa Rajab","doi":"10.4018/ijfsa.2020040102","DOIUrl":null,"url":null,"abstract":"Neuro-fuzzysystemsbasedonafuzzymodelproposedbyTakagi,SugenoandKangknownasthe TSK fuzzymodelprovideapowerfulmethod formodellinguncertainandhighlycomplexnonlinearsystems.TheinitialfuzzyrulebaseinTSKneuro-fuzzysystemsisusuallyobtainedusing datadrivenapproaches.Thisprocessinducesredundancyintothesystembyaddingredundantfuzzy rulesandfuzzysets.Thisincreasescomplexitywhichadverselyaffectsgeneralizationcapabilityand transparencyofthefuzzymodelbeingdesigned.Inthisarticle,theauthorsexplorethepotentialof TSKfuzzymodelling indevelopingcomparatively interpretableneuro-fuzzysystemswithbetter generalizationcapabilityintermsofhigherapproximationaccuracy.Theapproachisbasedonthree phases,thefirstphasedealswithautomaticdatadrivenrulebaseinductionfollowedbyrulebase simplificationphase.Rulebasesimplificationusessimilarityanalysistoremovesimilarfuzzysets andresultingredundantfuzzyrulesfromtherulebase,therebysimplifyingtheneuro-fuzzymodel. Duringthethirdphase,theparametersofmembershipfunctionsarefine-tunedusingaconstrained hybridlearningtechnique.Thelearningprocessisconstrainedwhichpreventsuncheckedupdatesto theparameterssothatahighlycomplexrulebasedoesnotemergeattheendofmodeloptimization phase.Anempiricalinvestigationofthismethodologyisdonebyapplicationofthisapproachtotwo well-knownnon-linearbenchmark forecastingproblemsanda real-world stockprice forecasting problem.The results indicate that rulebase simplificationusinga similarity analysis effectively removesredundancyfromthesystemwhichimprovesinterpretability.Theremovalofredundancy alsoincreasedthegeneralizationcapabilityofthesystemmeasuredintermsofincreasedforecasting accuracy. For all the three forecasting problems the proposed neuro-fuzzy system demonstrated betteraccuracy-interpretabilitytradeoffascomparedtotwowell-knownTSKneuro-fuzzymodels forfunctionapproximation. rule Base Simplification and Constrained Learning for Interpretability in TSK Neuro-Fuzzy Modelling","PeriodicalId":233724,"journal":{"name":"Int. J. Fuzzy Syst. Appl.","volume":"292 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2020-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"3","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Int. J. Fuzzy Syst. Appl.","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.4018/ijfsa.2020040102","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}

引用次数: 3

Abstract

Neuro-fuzzysystemsbasedonafuzzymodelproposedbyTakagi,SugenoandKangknownasthe TSK fuzzymodelprovideapowerfulmethod formodellinguncertainandhighlycomplexnonlinearsystems.TheinitialfuzzyrulebaseinTSKneuro-fuzzysystemsisusuallyobtainedusing datadrivenapproaches.Thisprocessinducesredundancyintothesystembyaddingredundantfuzzy rulesandfuzzysets.Thisincreasescomplexitywhichadverselyaffectsgeneralizationcapabilityand transparencyofthefuzzymodelbeingdesigned.Inthisarticle,theauthorsexplorethepotentialof TSKfuzzymodelling indevelopingcomparatively interpretableneuro-fuzzysystemswithbetter generalizationcapabilityintermsofhigherapproximationaccuracy.Theapproachisbasedonthree phases,thefirstphasedealswithautomaticdatadrivenrulebaseinductionfollowedbyrulebase simplificationphase.Rulebasesimplificationusessimilarityanalysistoremovesimilarfuzzysets andresultingredundantfuzzyrulesfromtherulebase,therebysimplifyingtheneuro-fuzzymodel. Duringthethirdphase,theparametersofmembershipfunctionsarefine-tunedusingaconstrained hybridlearningtechnique.Thelearningprocessisconstrainedwhichpreventsuncheckedupdatesto theparameterssothatahighlycomplexrulebasedoesnotemergeattheendofmodeloptimization phase.Anempiricalinvestigationofthismethodologyisdonebyapplicationofthisapproachtotwo well-knownnon-linearbenchmark forecastingproblemsanda real-world stockprice forecasting problem.The results indicate that rulebase simplificationusinga similarity analysis effectively removesredundancyfromthesystemwhichimprovesinterpretability.Theremovalofredundancy alsoincreasedthegeneralizationcapabilityofthesystemmeasuredintermsofincreasedforecasting accuracy. For all the three forecasting problems the proposed neuro-fuzzy system demonstrated betteraccuracy-interpretabilitytradeoffascomparedtotwowell-knownTSKneuro-fuzzymodels forfunctionapproximation. rule Base Simplification and Constrained Learning for Interpretability in TSK Neuro-Fuzzy Modelling

查看原文本刊更多论文

TSK神经模糊建模中规则库简化和可解释性约束学习

Neuro-fuzzysystemsbasedonafuzzymodelproposedbyTakagi，SugenoandKangknownasthe tsk_ fuzzymodelprovideapowerfulmethod formodellinguncertainandhighlycomplexnonlinearsystems。TheinitialfuzzyrulebaseinTSKneuro-fuzzysystemsisusuallyobtainedusing datadrivenapproaches。Thisprocessinducesredundancyintothesystembyaddingredundantfuzzy rulesandfuzzysets。Thisincreasescomplexitywhichadverselyaffectsgeneralizationcapabilityand transparencyofthefuzzymodelbeingdesigned。Inthisarticle，theauthorsexplorethepotentialof TSKfuzzymodelling indevelopingcomparatively interpretableneuro-fuzzysystemswithbetter generalizationcapabilityintermsofhigherapproximationaccuracy。Theapproachisbasedonthree阶段，thefirstphasedealswithautomaticdatadrivenrulebaseinductionfollowedbyrulebase simplificationphase。Rulebasesimplificationusessimilarityanalysistoremovesimilarfuzzysets andresultingredundantfuzzyrulesfromtherulebase，therebysimplifyingtheneuro-fuzzymodel。Duringthethirdphase，theparametersofmembershipfunctionsarefine-tunedusingaconstrained hybridlearningtechnique。Thelearningprocessisconstrainedwhichpreventsuncheckedupdatesto theparameterssothatahighlycomplexrulebasedoesnotemergeattheendofmodeloptimization阶段。Anempiricalinvestigationofthismethodologyisdonebyapplicationofthisapproachtotwo well-knownnon-linearbenchmark forecastingproblemsanda现实世界的stockprice预测问题。The结果表明，rulebase simplificationusinga相似度分析是有效的。removesredundancyfromthesystemwhichimprovesinterpretabilityTheremovalofredundancy alsoincreasedthegeneralizationcapabilityofthesystemmeasuredintermsofincreasedforecasting准确性。对于所有这三个预测问题，我们都演示了所提出的神经模糊系统betteraccuracy-interpretabilitytradeoffascomparedtotwowell-knownTSKneuro-fuzzymodels forfunctionapproximation。基于TSK神经模糊建模的规则库简化和可解释性约束学习

本文章由计算机程序翻译，如有差异，请以英文原文为准。

求助全文

约1分钟内获得全文求助全文

来源期刊

Int. J. Fuzzy Syst. Appl.

自引率

0.00%

发文量